Audio augmentation for speech recognition

By Tom Ko, Vijayaditya Peddinti, Daniel Povey, Sanjeev Khudanpur. 2015

1,266 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Speech & Audio, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Speech & Audio

In Signal Processing

Citations in the corpus

Do Not Have Enough Data? Deep Learning to the Rescue! (2019, neutral): “…apply transformations on existing training examples, such as cropping, padding, ﬂip-ping, and shifting along time and space dimensions, as these transformations are usually class preserving (Krizhevsky, Sutskever, and Hinton 2012; Cui, Goel, and Kingsbury 2015; Ko et al. 2015; Szegedy et al. 2015).”
FLEURS: FEW-Shot Learning Evaluation of Universal Representations of Speech (2022, background): “SpecAugment [23], speed perturbation [24], simulated reverberation [25], etc.”
SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition (2019, methodology): “Speed perturbation has been applied on raw audio for LVSCR tasks in [13].”
TED-LIUM 3: twice as much data and corpus repartition for experiments on speaker adaptation (2018, background): “…in red, which is similar to the Greedy one, but in which each training audio samples is randomly perturbed in gain and tempo for each iteration [4]; – the Beam+augmentation conﬁguration, in brown, achieved by applying a language model through a beam search decoding on the top of the neural…”
Deep Speech 2 : End-to-End Speech Recognition in English and Mandarin (2015, methodology): “. Techniques used for data augmentation in speech range from simple noise addition [26] to complex perturbations such as simulating changes to the vocal tract length and rate of speech of the speaker [31, 35]. Existing speech systems can also be used to bootstrap new data collection. In one approach”
You Do Not Need More Data: Improving End-To-End Speech Recognition by Text-To-Speech Data Augmentation (2020, methodology): “It also has been shown that widely used augmentation techniques, specifically speed perturbation [11] and SpecAugment [12], stack effectively with TTS augmentation [13].”
Personalization of End-to-End Speech Recognition on Mobile Devices for Named Entities (2019, background): “On the other hand, regularization [21, 22] and data augmentation [23, 24] techniques can also be used to prevent overfitting when training on a small amount of data.”
Unsupervised Adversarial Invariance (2018, methodology): “to learn models robust to such expected yet unobserved or infrequently seen (during training) variations is data augmentation through synthetic generation using methods ranging from simple operations [10] like rotation and translation to Generative Adversarial Networks [1, 8] for synthesis of more r”
Libriheavy: A 50,000 Hours ASR Corpus with Punctuation Casing and Context (2023, methodology): “SpecAugment [22] and on-the-fly Speed perturbation [23] are also applied to augment the training data.”
Jasper: An End-to-End Convolutional Neural Acoustic Model (2019, methodology): “At training time, we use 3-fold speed perturbation with fixed +/-10% [32] for LibriSpeech.”
Letter-Based Speech Recognition with Gated ConvNets (2017, background): “(Panayotov et al., 2015) 5.5 14.0 (Amodei et al., 2016) 5.3 13.3 (Peddinti et al., 2015b) 4.8 - (Povey et al., 2016) 4.3 - (Ko et al., 2015) - 12.5 this paper 4.8 14.5 this paper ( no decoder ) 6.7 20.8 Using a single GPU (and no utterance batching), our High Dropout Gated ConvNet goes over the…”
A Comparative Study on Transformer vs RNN in Speech Applications (2019, methodology): “, feature extraction, SentencePiece [19], language modeling, decoding, data augmentation [26], [27]) except for the training stage to their optimum in the existing RNN-based system.”
Specaugment on Large Scale Datasets (2019, background): “Speed perturbation [11] has been an integral part of augmentation of speech data.”
Transformer-Based Acoustic Modeling for Hybrid Speech Recognition (2019, methodology): “Speed perturbation [34] and SpecAug-ment [10] (LD policy without time warping) are used.”
Data Augmenting Contrastive Learning of Speech Representations in the Time Domain (2020, positive): “Basic time-domain augmentations modify the sampling rate of the input by a small factor ( ± 10% ), which changes both the duration and pitch [15].”
EDA: Easy Data Augmentation Techniques for Boosting Performance on Text Classification Tasks (2019, methodology): “Automatic data augmentation is commonly used in vision (Simard et al., 1998; Szegedy et al., 2014; Krizhevsky et al., 2017) and speech (Cui et al., 2015; Ko et al., 2015) and can help train more robust models, particularly when using smaller datasets.”
Conditional BERT Contextual Augmentation (2018, background): “Recent years have witnessed great success in applying data augmentation in the ﬁeld of speech area(Jaitly and Hinton, 2013; Ko et al., 2015) and computer vision(Krizhevsky et al., 2012; Simard et al., 1998; Szegedy et al., 2015).”
Contrastive Semi-Supervised Learning for ASR (2021, methodology): “Our input speech features are 80 dimensional, speed perturbed [28] mel-scale log filterbank coefficients computed every 10ms over 25ms windows.”
Purely Sequence-Trained Neural Networks for ASR Based on Lattice-Free MMI (2016, methodology): “We use the speed-perturbation technique ([16]) for 3-fold data augmentation; and iVectors to perform instantaneous adaptation of the neural network ([17]).”